Effective Matrix-Vector Multiplication Computations on Account of Physically Based Matrix Distribution

This paper considers a parallel algorithm for Cholesky factorization on account of PBMD. Compared with the traditional data distribution for matrix-vector multiplication computation, the algorithm proposed incurs strictly less communication overhead and improves the load balancing. Experimental results demonstrate that the PBMD based matrix-vector multiplication computation makes the implementation of Cholesky factorization easier and enhances performance effectively.